MAIster documentation

MAIster is a self-hosted control plane where people and AI agents deliver software through shared, repeatable processes. It connects tasks, isolated workspaces, coding-agent sessions, human decisions, evidence, costs, review, and promotion without making team members babysit terminals.

​

Start with the product

If MAIster is new to you, follow the visual tour before changing configuration:

• Application map explains the portfolio, Inbox, global administration, and the path into a project.
• Personal librarian helps you find, create, clarify, and launch tasks in one private conversation, then follow results.
• Project workspace covers every project tab, from the board and activity to packages, integrations, Brain, and settings.
• Repository and project packages shows how to browse source and attach upstream or locally cut packages.
• Kanban board, tasks, and Runs follows work from the queue into one or more execution attempts.
• Run history and the Run workspace covers the ledger, graph, node attempts, files, evidence, and timeline.

Then install a local instance and register a repository.

​

Configure the platform

​

Build delivery processes

Flow Studio edits installed processes as a graph, forks package content, and can use an AI assistant to propose bounded changes. Learn how to open and use the assistant, then continue with the node catalog and structured results and Run context to pass typed data between graph steps.

​

Inspect quality and economics

• Compare Runs across Flows, coding agents, models, capabilities, objective tools, AI judges, and human verdicts.
• Use Observatory to attribute tokens and time to tasks, stages, runners, and models.
• Follow review, rework, and human takeover from production artifacts to agent correction or local editing and back into Flow validation.
• Configure Project Brain, embeddings, and platform agents for indexed sources and governed memory.

​

For AI agents

Every page has a unique title, explicit prerequisites, success or failure signals, and stable links. The self-hosted build exposes source Markdown under /markdown, plus /llms.txt and /llms-full.txt. A Mint-hosted deployment can also expose a documentation MCP endpoint at /mcp. Prefer the page closest to the requested operation. Do not skip declared human approval, access, evidence, or promotion boundaries.

​

Current operating boundary

The current production topology keeps the web control plane and execution supervisor on one trusted host with access to the same repositories and worktrees; Postgres stores the durable ledger. The architecture has stable host identities and ownership epochs for further multi-host work, but the public operation guide does not promise transparent mid-session migration.